Discussions on Creating Schemas

"person" is another example of this.  There is a "person" type in freebase.  There is also a "person" topic (which references the wikipedia article on
person") 

 The way this is currently handled is: the topic and the type are two different objects in freebase.  The topic is marked as an instance of a type called: "type/domain equivalent topic". This may seem a little odd but I think it's a way of clearly separating types from topics.

 Whether a topic, such as "Talaria" merits having an equivalent type is up for debate. We're talking about a mythical, arguably fictional thing that is probably only talked about as a concept, i.e. no one talks about *that* particular Talaria and how fine it was.  What would be an "instance" of talaria? I don't think there's any practical value of having such a type.

So, by this criteria, having a type "Sandal" is practical: instances could include consumer products (e.g. "The Birkenstock Arizona") And the "Sandal" type could be included by another type "Huaraches" referring to the subset of sandals found mostly in Mexico.

I'm sorry, I'm not helping with the root question "how could we get WordNet data in freebase"

Maybe alternatively create a 'WordNet entry' type, add that to the topics that are listed in WordNet and add such fields as 'hypernyms', 'hyponyms', 'meronyms', et cetera.

Or is that too specific to one data source?

I think that Freebase’s internationalization makes a difference here; to me, words should be distinct from concepts, but connected. For instance, the topic for dog has names in multiple languages (page down in the Explore view), and not everything that WordNet will have to say about “dog” will apply to the organism. I would make a type for Word (with a “Language” property) or for English Word to use for WordNet topics, and link the Word instances with the topics that they describe in English.

The hypernym/hyponym relationships apply not only to the words but to the topics themselves (e.g., that “dog” is a hyponym of “animal” implies that a dog is a kind of animal), and that information could be applied to the topics themselves, but the words have, IMO, an independent existence.

Well, sure, “dog” the animal and ”dog” to secure a hatch have different WordNet entries. But there are still things that one can say about “dog,” the word referring to the animal, that aren’t true of “chien” (and vice versa), while that is not the case for dog/chien/hund/etc. as a concept, especially etymological observations.

This is a project I find exciting, but it (like most interesting projects) is hard to do right. 

Stumbled over this old conversation, and I'd like to revive it as I've just created a symset type which has properties for hypernym and hyponyms.

I think this symset type might be suitable for filling with a WordNet import? 

I agree with brendan that creating a type for each WordNet symset would generate a lot of fairly useless bucket types.  With the symset type each topic in freebase would relate to a unique Wordnet Symset.  Unfortunately, the downfalls of this symset type are:

• we'd have to fill out the hypernym/hyponym properties of symset (duplication with existing types)
• having a lot of topics typed as "synonym set"
• Need to link a symset to the words/synonyms used in a language to describe it.

The latter is the biggest problem as a symset can have multiple words in each language, and in reverse a word can have multiple meanings - i.e. there's a many-to-many relationship between a synonym/word and the synonym set.

Also I wanted to allow a word to be split into its component morphemes.  So decided to create a CVT for the synonym property and the word being text input.  (essentially the text would be a duplicate of the /common/topic/alias property)

The word "Rat" is my initial example - I've a topic for the animal symset, and another topic for the informer symset, all linked by the word CVT. 

Again, I'm not convinced there's a general enough use case here to make this worthwhile. Certainly, you'll discover a lot of (IMO, justified) resistance to encoding special cases like this in the GUI.

If you can explain what you're trying to do here, people may be able to suggest ways to help you acheive that in a different way.

I assume what you want to do is to see every award that has a category for best actor.

I think the reason for having the entire title in the name is to easily differentiate the Academy Award for best actor from my local high school's drama class prize giving; or indeed any other award category.

Having a separate topic for each award category also allows us to note specific details for the category e.g. first awarded and last awarded dates.  A referential table (aka Compound Value Type, CVT) wouldn't easily allow us to do this.

However, if you'd like to group all the best actor award categories, I'd recommend creating a new type and apply it to the relevant award category topics.  This type could have a property called "category", which would expect topics such as "best actor", "best actress" etc..  That way you could link up all "best actor" award categories.

In the schema editor, you should be able to edit a property and the delete button is shown on the top right of the edit box.

You can always hide the property, preventing information from being added to it, this can also be selected from the edit box for the property in the schema editor. 

I've created a quick query for all /type/property created by /user/polojan http://tinyurl.com/m3c96a

 and another for orphaned properties by /user/polojan (orphaned as in the expected type is not given) - http://www.freebase.com/app/queryeditor/?q=[{%20"type":%20"/type/property",%20"id":%20null,%20"creator":%20"/user/polojan",%20"expected_type":%20{%20"id":%20null,%20"optional":%20"forbidden"%20},%20"schema":%20{%20"id":%20null,%20"name":%20null%20}%20}]

Thank you for the MQL. There are the properties that I was talking about, however, when I open them it says

... "The page you were looking for doesn't exist."

So I guess this is only a caching matter and they will dissapear completely after a while.

In the interest of speed, the client deletes the minimal amount of links to allow the user to resuse the keys when creating types and properties.  The proper way to clean up is to remove any links from properties, then delete the properties, remove any instances typed as the type and then delete the type.

 This shouldn't be an issue for much longer; we are working on a backend process that will clean up these deleted types/properties on a regular basis.

 If you have some properties/types that you would like cleaned up, drop me an email with your ids and I'll take care of it.

Question 1: There are a slew of properties called "comment", and only 10 results are returned in autocomplete, so yours isn't showing up. You can pass in the id of the property (/base/facetedbrowsing/faceted_browsing_result_view_support/comment), though, and it will show up.

Create the table that you want with the right filters and right columns displayed, then click on the little gear icon on the top and select "Download as a file."

Don't forget that the Terms of Service require attribution back to Metaweb/Freebase.

Not sure what you mean about the heirarchical vs relational, but the database isn't either really.  If you're talking about the information displayed on a topic page in Label: Value form, you can get all that to come out in table views as well with the appropriate amount fiddling about the the table settings (click Edit View and add additional types/columns).

For data that's missing, feel free to add it if you've got a good source that's license compatible. 

Does that work for you?  It didn't for me.

I ended up struggling through the explore2 view and raw MQL to create what I needed.

Thanks. 

p.s.  The topic should, of course, be "Out of date." That's what happens when you stare at MQL for too long!

Thanks jeff. I found the help docs.

 I've created a new Lineage Member type, which currently has "student of" and "students" properties.   May I suggest adding it to the Influence Commons schema?  Here are the sample topics that I've used the type with: http://www.freebase.com/view/user/flyingzumwalt/default_domain/lineage_member

Note: these links might break when I move the type out of my default domain. 

Pleased to oblige.

Example: The Mahamudra Lineage (Kagyu School of Tibetan Buddhism)

The Mahamudra Lineage is well defined.  Tilopa taught Naropa, who then taught Marpa.  Marpa returned to Tibet and taught Milarepa.  The lineage then continues and branches - Gampopa, Dusum Khyenpa, etc.  and can be traced all the way to living lineage holders today (over 1000 years of oral transmission).  The resulting branches of transmission formed the 4 major and 8 lesser lineages of the Kagyu school.

This stuff is important if you want to make sense of any oral tradition, especially when you look at events like the Rimey movement where a few very influential yogis spent their lives rescuing and consolidating dying lineages.  The resulting texts, which attempt to preserve the unique flavor of each lineage of instruction, have become some of the most important and most frequently referenced texts in modern Tibetan Buddhism.

In light of that, take a look at Milarepa's influence node.  He was influenced by both the Buddha and Marpa.  There's no way to distinguish that Marpa & Milarepa had a very significant personal relationship of instruction while the Buddha was simply an influence.

You might be tempted to say that it is sufficient to mark people as contemporaries (either explicitly or by means of birth/death dates).  This would fail though.  Proof: Anyone using freebase right now has been influenced by Tim Berners-Lee and we are all his contemporaries, but very few of us have actually been his students.  There is no way for me to find out who his actual students are.  The existing ontology is insufficient.

I think that highlights precisely the issue that Jeff was talking about.  Did Gampopa (Thinkbase link) really not "influence" some of his students or did someone just forget to do the redundant data entry (made necessary by the denormalization) in the "influence node" part of the graph?

Having said that, the Academic type that Jeff pointed to has precisely the same problem. Perhaps the Influencer/Influencer link needs to be qualified by type or strength or ...

One of the things of this type that I think it would be cool to be able to include is the Math Genealogy project http://www.genealogy.ams.org/

I agree that Written Work->Online-Versions is weak and should probably go away, but rather than shoe-horning new properties onto an unrelated type like Book Edition, I'd consider creating a whole new type called Online Edition or something.  That way you could drop all the physical world characteristics and include the important virtual world characteristics.

This would allow pure digital editions for works that were never printed on a dead tree, would allow for mutliple different digital versions (e.g. image only vs machine OCR vs OCR+proofreading, Microsft vs Google books, etc),  and would also allow you to link the different editions together in the same way that you would a translation.

I made some first efforts:

 http://jewlib.freebase.com/type/schema/base/jewlib/book_edition

Of course, it might be better to have a common type of this.

I see a practical problem arise: You have to relate each type you use to a more general one.  and  you have 4 types to describe maybe just one  online book edition (if you want to do it correctly), like:

http://jewlib.freebase.com/view/guid/9202a8c04000641f800000000b49ad03 

 (I tried to add the editions to Book type, but I got Error messages)

 What do you think. I mean, almost every book will have a online-edition in future. So we should consider to extend the book-edition type? What do you think, to keep it uncomplicate?

I will also post that to the mailing list,

 best regards,

yours

Frank

And: "physical world characteristics" of books are congruent with most of the characteristic of a digitized book. You only have to add information, like repository and the URL, of course. So you have the opportunity to link both items closer.

 yours

Frank

Hi,

Great to see interest in digitized books.

I think it would make sense to extend/modify the existing Book Edition type so that it captures all Frank's desired properties, instead of creating a new type.

First of all, Book Edition was desired to include digitized editions as well. Book Eidition has a property for Book Binding, which has instances such for E-Book and Portable Document Format, among other things It also captures all the generic properties any book edition would have, such as publiser, year, etc. For ease of use, having one type that covers every edition of a written work means users only have to make one query to get all edition info: from hardback, soft-cover, to large print and mass market edition, from books on tape and CD, to digital downloads in various formats.

If we can work to include desirable properties for digitized books in Book Edition, would that suffice?

Hi there,

 it would be great if you can work to include desirable properties for digitized book in Book Edition!

I thought about my suggested properties for a new type Book edition again. Below you will find my suggestions:

1) URL (is it possible to get a Link into the type section? at the moment, links are only shown on the right side, section weblinks)

2) Digitized Book Contributor (in general a library, but there are some private projects, digitizing public domain books)

3) Available through (name of the internet library, e.g. archive.org, books.google.com etc.)

4) Digitizing Sponsor (in general Google or MSN, sometimes national research foundations like DFG, see. http://www.judaica-frankfurt.de/)

5) Scanning Center (don't know if it is important right now, but mabye it could be helpful to link to pages of scanning centers in the framework of freebase to get to know something about the quality of the documents)

6) Direct Link available (yes/no; sometimes you can only link a book collection)

 best regards and thank you very much

yours

Frank

These are great suggestions.  One thing we need to consider is that there are many kinds of online materials, not just digitized books, and we should allow for links to and data about other online materials, whether at Project Gutenberg or an author's or fan's website. It's possible that the "online versions" property could be used for simple links, where the metadata isn't important (or is essentially nonexistant), such as links to stories or articles published in an online magazine.

One question about this is whether it's important to know which edition of a work has been digitized, or whether it's enough to know only that an edition has been digitized. For some books, there are significant differences between editions, although for most the differences will be trivial. But as a researcher for whom the first publication is an important consideration, being able to tell which edition(s) have been digitized would potentially be useful.  We should also consider that different digitizers are going to scan many of the same books, whether in different or the same editions.

If we do limit this to books, what just occurred to me now would be to add the suggested properties to a CVT that is attached to Book Edition; this would allow for multiple digitizations of the same edition to be recorded easily. The alternative seems to be to treat each digitization as a new edition, which would also be possible, but I could imagine the data-entry getting somewhat confusing in that instance.

One other thing that might delay full implementation of this (although it's not a reason to stop thinking about the model) is that there may be a refactoring of the Book Edition type happening soon, based on some dataloads that we're considering.  I'll post more about that when there are any specifics.

I can see the value in have a single type to hold the information about published editions, so I withdraw my suggestion that online editions be separate from paper editions.

As for Jeff's questions, I believe you absolutely need to be able to tell which editions were derived from which and you need to support multiple derived versions (whether they be translations, digitizations, or transcriptions) of any given edition.  Each source/derived pair needs information specific to that particular transformation (name of the translators, organization doing the digitizing, etc), although that information can just as easily be associated with the resulting edition as with the association between the two editions.

Hi there,

I am absolutely with you: "you need to be able to tell which edition were derived from..."

 I just want to add, that there is also a need to do same changes for periodicals, like journals and newspapers. (See, e.g. http://www.compactmemory.de/):

http://www.freebase.com/type/schema/book/journal?domain=%2Fbook

http://www.freebase.com/type/schema/book/newspaper?domain=%2Fbook

A lot of online journals and newspapers are not fully digitized. Often you only will find single issues or volumes. Maybe you have to include apropriate fields (in addition to the suggestions for "Book Edition"), like:

 available from "date [issue no., Year] " to "date [issue no., year] "

We should also consider, if there is a need to mark publishing periods with different editors [and alternating periodical names]. This might be very helpful for researchers.

 yours,

 Frank

Hi there,

 I found the "Periodical type. Maybe you can also include here properties which include information on online versions.

best regards,

f.

The Periodical type definitly would be the right place for data about digital versions.  One question is whether there should be different properties for third-party digitizers/archives (Google, JSTOR, etc.) and for a periodical's own online archives.  But regardles, I agree that having (possibly multiple) date ranges per archive for a periodical would make the most sense, rather than trying to link to individual issues.

Thinking about books, I'm starting to think that digitized versions of book editions hosted in an archive (Google books, OpenLibrary, etc.) might need to be modeled differently than commercially availble ebooks. That is, an ebook is just a type of edition -- it's got a publisher, and pub date, possibly an ISBN, etc., but is not necessarily purporting to be a faithful reproduction of a print edition (and may indeed have undergone editing in the publication process that changes the text with regard to a print edition).

The other issue is versions of books that are just hanging around the internet because someone (the author, publisher, a fan) put them there. I'm not really sure if there's a model for these, and they're partly what the "online versions" link was intended for.

I like the book-type on the sandbox!

You can’t, for a couple of reasons. First, if you add a previously-unknown organizer, what type should Freebase make it? Second, if you want to make a reverse property for organizer (organizations organized, perhaps), what type should it be on?

The thing to do in this case is to make a new type, probably called Organizer. That way the Person or Organization that did the organizing can have a reverse property easily.

I see.

It is a common case to have union of types as the range for a property. I think that creating new types in this way may result in many types.Anyway, thank you very much.

Are you perhaps trying to do a GET rather than a POST with your servlet? A 405 is usually triggered by using an incorrect method for a Web service, and only POST is allowed for mqlWrite. Since your query works through the Query Editor, this is my first guess. You might also want to take a look at Chapter 6 of the MQL Reference Guide, which covers the mqlWrite service:

http://mql.freebaseapps.com/ch06.html

But im using the "doPost" method in a servlet created by Eclipse? Are there example MQLWrite java codes provided by Freebase?

Thanks

I've created a new Energy unit, "Ton of TNT" that you can use. (I added kiloton and megaton of TNT for good measure.)

One other thing is that, in order to assign a unit, you'll have to change the expected type from floating point range to a plain old float, or create a new range CVT so that you can assign the units to the float properties. (Because the float range type is generic, units can't be assigned to properties it has.)

@spencermountain - you are correct, an explosive charge is measured in 'tonnes of TNT' or 'kilogram of TNT' with regards to blast effect on structures  - for example see Table 2.2 on Page 69 of http://books.google.co.uk/books?id=g-L1O4YszQgC.

@jeff - could 'Kilogram of TNT' be added?

alright!

while we're at it, how bout a megawatt datatype for  powerplant?

hey, in one sentence or less, (joke) are freebases datatypes 'semantic? does killoton understand its 1000 times ton?or does each programmer do that stuff adhoc?

cheers guys. thanks.

Gotcher megawatt right here.

Now, whether Freebase's types understand anything or not is an open question. As far as we know, the graph hasn't achieved full sapience yet, but it could just be playing dumb.

If you look at the megawatt type, though, you can see that it has a property "Power in Watts" which can be used to do conversions between units that measure the same dimension (power, energy, length, etc.). There's nothing in Freebase to do this for you at the moment, but the structure is there if anyone wants to make use of it.

I'm a bit ofo f fiend for images and this base is very visually appealing.  I started loading images last night - finished off the missing ones from the Atomics and the Dark Horse covers.  I'll be doing more for sure!

j

Jon, I agree and have removed the type and fixed the incoming topics.

However, I can see where it came from, the architect Eugene Viollet-le-Duc was primarily only involved in restoration of old buildings, thus he has been noted, perhaps incorrectly, as the architect of the building - nothing distinguishing him from the original architect.

 For the Building Restoration topic, I've quickly crafted together a Project Phase type in the Engineering Base, and any comments would be helpful.

However, this sidesteps the re-occuring constraint with the structure type - the structure type does not take into account that a structure can have multiple architects and engineers, each involved at different times on different phases of the structure's lifetime (e.g. Initial design, compared to a later restoration).  Is there a quick fix to this, that doesn't add too much complexity to the structure type?

Just address a particular facet of this: When a building is first built, things are a more straight forward: there is an architect (multiple individual architects) and/or a firm(s).  The current "structure" schema supports this information.

 What to do about subsequent renovations,  I'm not sure, It's somewhat of an edge case, so some provisional schema seems prudent at this point. 

 In the case where an architect is essentially re-designing a building, there may be some things that can be done. There are numerous cases of old buildings where a key part was kept while a completely new building, by a new architect, in a different style was built up around it.   Is that a new building?  A simple way of approaching this case is: for the original structure topic. fill in the "Destroyed By" -> "Renovation" and the "Destruction Date".  Then create a new topic representing the "new" building.

Just to add: the two cases I've given are ones where I am asserting, is this "entity" really the same entity, now that a renovation has occured?

 Here's another example: The Contemporary Jewish Museum

 (without addressing the conflation of the museum and it's main building!)

It seem confusing to list Willis Polk as the architect of this building (he designed the original building that this new design was grafted on to).   I'd prefer to have another topic with it's own picture and details of it's design (and the assertions that this building, for all intents and purposes, no longer exists)

Does that make sense?

@brendan - perfect sense.  Where buildings have been demolished, or majorly regutted and rebuilt it would be better to have two different topics.   The Tay Bridge which was destroyed in an infamous disasterand completely rebuilt is another example where two distinct structures share one topic as they have a common location and name.

@skud - I really like your project schema, I think it solves all my issues here.  With regards to structures I would assume that the project topic would be named, for example, "Design and Construction of the Empire State Building" and would link to the "Empire State Building" topic through the Project focus property?

I had wrestled with doing this with my 'Engineering Project' type, but I didn't want to start atomising topics (i.e. splitting a topic into a 'Construction of x', 'Design of x' and so on) - and instead settled on a CVT called 'Engineers Role'.

If there is a consensus amongst the Architecture domain that the project schema is the way forward, I am happy to move my Engineering Project data to it and delete the superseded types in the Engineering domain.

@sprocket: You are absolutely right about the Empire State Building example.

I showed Brendan the Project schema yesterday f2f at the office, and he liked it.  We talked about promoting it and making types like /architecture/structure and /computer/software include the Project Focus type.  Brendan seemed fairly positive about the idea of deprecating and hiding the project-related fields on /architecture/structure and gardening the contents of them out to the project types.  The problem with that would be that we'd break some saved views and apps, but then, that's always the problem with schema changes.  However, the number of people using stuff from /architecture is fairly contained and we can probably manage it.

 If you wanted to make use of the Projects schema for your Engineering stuff, that would be a great test run of it.  I have to admit I was thinking fairly abstractly and didn't really have any good examples to enter data for (though I did try out the Manhattan Project, as a potential edge-case, and it seemed to work OK).  So it would be great to have some real data using it so we can kick the tyres a bit before promoting.

I've added some structures data to the project type.  I chose a particularly complex example, the temporary skylon structure and its associated projects for the 1951 Festival of Britain, as a starting point.

 The type seems to work very well (except that the project focus doesn't return link to project) , and I was particularly pleased at being able to note the dignitary who laid the foundation stone of the Royal Festival Hall!

jeff - good point.  I don't think there is enough data on this particular topic to warrant seperating the design from the construction.  So I've renamed it 'Initial design and construction of the Royal Festival Hall' which describes my intent slightly better.  

skud - I've also filled out the recent Renovation of the Royal Festival Hall, so now we have a project focus with two projects.

Thank you Faye.  I appreciate your tip!  I welcome any info that will better ensure correct entries.

 Have a great day!  Thanks again!

I think there's value in keeping some notion of Patent as a standard type (the description, for starters, would go away if we changed it into a CVT), but I agree that it should be internationalized.

The simplest thing might be to add a new property for "Patent-granting body" as you suggest, and remove or generalize the US-specific properties.  There are already properties for application date and issue date, but I think status sounds pretty good.

Can you give an example of what "commands" you had in mind?

As for your other questions, you should have a look at the Data Modeling Guide and the Beginner's Guide to Creating Schemas, these have a lot of information about how to develop the structures (or "schema") for representing knowledge in Freebase, and how to create relationships between properties and share information.

Hi mateolan!  I'm the owner of the Food base, and I'm a member of staff here.  In my team we often work on draft schema in bases, and then those get promoted into the commons.  So originally I created my Food base to work out some schema around Dishes and Ingredients and a few other things.  Some of these were promoted, and some weren't, but what you see hanging around in the Food base now is just kind of a working area for me.

 Also, the "based on" thing only came into existence around November '08.  Any bases created before then (actually, we called them user domains, not bases, at that time) weren't "based on" the Commons.  "Based on" occurs when you create a saved view using a type in the commons, and save it to your base.  That's only been around for a little while.

I agree.

This could be modelled by having the Transit Lines property have a Data Type of 'Transit Line Stops', a new type which would be a CVT.  Properties for 'Transit Line Stops' would be 'Transit Line', 'Transit Stop', 'Date Stop Added To Line' and 'Date Stop Removed From Line'.